Research on Chinese Chess Detection and Recognition Based on Convolutional Neural Network

Abstract

A deep learning method based on convolutional neural network is introduced into the human–machine game of Chinese chess to detect and recognite pieces, which uses the faster R-CNN target detection architecture and the GoogLeNet convolutional neural network. Based on the shape features of pieces, an anchor mechanism is designed to detect pieces. The method of extracting eight-direction features and rotating invariant HOG features are used to train the ResNet convolutional neural network, which can improve the problem of fewer types of training-intensive chess pieces, single rotating pictures, etc. Experimental results show that the proposed convolutional neural network for chess detection and recognition can greatly enhance the detection of environmental adaptability, robustness, and improve the detection speed and the recognition accuracy.